Modeling semantic trajectories including multiple viewpoints and explanatory factors: application to life trajectories David No¨el, Marl`ene Villanova-Oliver, J´erˆome Gensel, Pierre Le Qu´eau
To cite this version: David No¨el, Marl`ene Villanova-Oliver, J´erˆome Gensel, Pierre Le Qu´eau. Modeling semantic trajectories including multiple viewpoints and explanatory factors: application to life trajectories. ACM SIGSPATIAL International Workshop on Smart Cities and Urban Analytics (UrbanGIS) 2015, Nov 2015, Seattle, United States. Proceedings of the 1st International ACM SIGSPATIAL Workshop on Smart Cities and Urban Analytics (UrbanGIS), 2015.
HAL Id: hal-01212903 https://hal.archives-ouvertes.fr/hal-01212903 Submitted on 7 Oct 2015
HAL is a multi-disciplinary open access archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destin´ee au d´epˆot et `a la diffusion de documents scientifiques de niveau recherche, publi´es ou non, ´emanant des ´etablissements d’enseignement et de recherche fran¸cais ou ´etrangers, des laboratoires publics ou priv´es.
Modeling semantic trajectories including multiple viewpoints and explanatory factors: application to life trajectories. David Noël Univ. Grenoble Alpes, LIG F-38000 Grenoble, France +33476827211
[email protected]
Marlène Villanova-Oliver Univ. Grenoble Alpes, LIG F-38000 Grenoble, France +33476827280
[email protected]
Jérôme Gensel Univ. Grenoble Alpes, LIG F-38000 Grenoble, France +33476827280
[email protected]
Pierre Le Quéau Univ. Grenoble Alpes, PACTE F-38000 Grenoble, France +33476825920
[email protected]
ABSTRACT In this paper, we propose a semantic trajectory model handling complex trajectories. We consider that a trajectory is composed of multiple aspects which are all different viewpoints from which it can be observed. This model also allows for the representation of explanatory factors that influenced the episodes that compose a trajectory. We here deal with a particular type of trajectory: the life trajectories of individuals or citizens. The residential aspect in life trajectories is particularly relevant for decision makers and urban planning experts who try to understand the residential choices of inhabitants in order to adjust their decisions. Those choices depend on factors related to the home (type, size...), and to the surroundings (accessibility, amenities ...) but they can only be fully apprehended by taking into account the life circumstances of the individuals, along their family, or professional or even spare-time activities viewpoints.
Keywords Semantic trajectories, Residential trajectories, Geospatial lifelines, Individual lifeline
1. INTRODUCTION Urban planning experts and decision makers face many challenges related to residential choices of inhabitants of metropolitan areas. These choices are strongly influenced by the various phases individuals go through in their personal lives. Thus, in general, couples with children prefer to live away from the metropolitan centers while those latter, in general, attract a higher proportion of both newly independent and retired people [7]. This illustrates some of the most important problems that major cities have to solve. This exodus of working couples with (young) children to peri-urban areas makes it difficult for the cities of the peripheries to provide the necessary amenities (for instance, nurseries and
schools), but also increases the time workers need to commute from home to work, which generates pollution and traffic congestion. Similarly, in city centers, one witnesses the emergence of problems of cohabitation between two age categories with sometimes different lifestyles. The diversification of the population of city centers and the periphery of these cities is an important issue and it is essential for decision makers and urban planning experts to acquire a better knowledge of residential dynamics. In particular, in order to make their decision, they try to better understand what is at stake for an individual or a household in the process of choosing a residence. Reasons for residential choice can be provided from a comprehensive approach to life trajectories of individuals that takes into account many decisive aspects of their lives: familial, professional... The living conditions, especially what is related to the characteristics of the residence and its neighborhood, the lifestyle preference [23], as well as other elements linked to the economic conditions must also be considered [10]. In this work, we specifically aim to take into account how individuals feel impacted by these elements. Then, in order to study the residential trajectory of an individual (the succession of its residential choices in time and space), it should be considered as an integrated part of a broader vision of her/his life trajectory. This part is one of the multiple and interconnected viewpoints (already demonstrated in computer sciences by Thériault [18]) from which the life trajectory of an individual can be observed. Our goal is to provide a generic model able to support this complexity related to the multi-dimensional nature of a life trajectory and to explanatory factors that each viewpoint may encompass to explain its evolution or the evolution observed in another viewpoint. For example, in the case of life trajectories, a move (from the residential viewpoint), may be explained by a change from a professional viewpoint (career development, job loss, etc.). Beyond this trivial example, one issue is the
representation of the combination of several explanatory factors of various kinds (a change of job may have imposed a move, the beginning of a sense of insecurity may initiated it). Finally, another issue is the opportunity to represent, beyond the observed true facts (e.g. an observed true move), projects that contribute to an ideal of residential trajectory (e.g. a desire to settle down in a particular place, or in a particular type of home, etc.). In this work, we therefore provide a model handling the complexity and variety of trajectories by combining the concept of viewpoint with a time structuration relying on episodes and events. Our model includes also explanatory factors as they may be reported by individuals. This work uses life trajectories as an application case for the modeling and analytics of human mobility in the urban context. The rest of the paper is organized as follows: we introduce in section 2, the notion of life trajectory. We present in section 3 a brief state of the art on semantic trajectory modeling and life trajectory modeling. Then we propose in section 4 our semantic life trajectory model which integrates multiple viewpoints and explanatory factors. Section 5 describes how this model can be exploited. Finally, we conclude and give directions for future work in section 6.
2. REPRESENTING AND STUDYING LIFE TRAJECTORIES Figure 1 (adapted from Thériault [18]) shows a part of the life trajectory of an individual. The life trajectory is represented here as a succession of states which are characterized by values of different statuses or attributes describing this individual. These attributes can be grouped according to the different aspects of the life of a person (such as residential, family, work, hobbies, etc.). These different groups of attributes can therefore be considered as different trajectories targeting a particular theme. Therefore, our representation of life trajectory is multidimensional, each dimension corresponding to a "thematic" focus. Thus, for example, the residential trajectory is represented by statuses is owner, rent/loan, housing type and city (see Figure 1). Any status, and consequently any trajectory is intrinsically temporal, as related to a timeline that ordered its different states.
Figure 1: A part of a life trajectory (Adapted from Thériault [18])
In Thériault’s work (see Figure 1), the residential and the professional trajectories are spatiotemporal trajectories since they each contain an attribute describing spatial information: the city of residence, and the city of employment. Such an information could be considered at another spatial scale, such as a neighborhood for the residential trajectory or as x, y coordinates. Other aspects of the life of the individual may also contain a spatial component, and be useful for the study of residential trajectories, for example the location where a hobby is practiced. As underlined by the French sociologists Yves Grafmeyer and Jean-Yves Authier [14], the term “residential trajectory” “tends to suggest that successive positions in a given set is not simply by chance, but rather are linked together by an intelligible order.” As a matter of fact a residential trajectory is not just a series of spatiotemporal positions. Part of our objective is to make it possible to understand the meaning of residential trajectories by bringing to light the underlying reasons why individuals move in a given space, at a specific moment of their lives. The degree of control of an individual on her/his trajectory differs, and it seems to be appropriate to highlight the conditions under which each individual makes her/his own choices. More broadly, we want to elicit the explanatory factors or combinations of factors that have shaped a given trajectory. Thus, somehow what we seek to bring out here is close to the notion of career such as it has been defined by Howard Becker [6]. This concept “refers to the sequence of movements from one position to another” and includes “both objective facts of social structure and changes in the viewpoints, motivations, and desires of the individual”. In the case of a residential trajectory, these factors are first to be found in the context surrounding the life trajectory of an individual. There are of course some internal factors which are directly linked to the specific characteristics of one or more viewpoints of this life trajectory. But, there are also some explanatory factors which can be external and so exogenous to the considered life trajectory. Among such external factors are, for instance, the housing market or change in the surroundings of the housing, etc... These external factors have nevertheless to be considered in correlation with the life trajectory. For instance, the fact that housing market prices are rising may have an impact depending on the salary of the individual/household and thus on her/his professional trajectory. In this work, we are particularly interested in those external factors which may be considered as plausible explanations, whether they are explicitly mentioned by the individuals or not. Representing life trajectory is complex because it is composed of many descriptive attributes which may refer to different aspects of the individual life. The structuration using viewpoints aims to facilitate the overall comprehension and to offer a better readability of some salient characteristics attached to a particular thematic focus. The representation of a life trajectory is highly dependent on the domain of the study and on its purpose. For a study in urban development, the focus is on the attributes describing the residential trajectory, but explanatory factors can potentially be found in other foci. The study of life trajectories requires the use of suitable referential for considering the various parts of the trajectory separately but also in their interrelationships. Thus, a relevant modeling will enable the integration of explanatory factors of the trajectory, whether internal or external.
3. RELATED WORK 3.1 Modeling semantic trajectories Many semantic trajectory models have been proposed in recent years, offering different solutions to enrich space-time trajectories. Most of these semantic trajectory models are based on different kinds of segmentation proposed for spatio-temporal trajectories. Indeed, following Hägerstraand and the "time-geography" [15], researchers have modelled the spatio-temporal trajectories through periods of trips and periods of activities [24], or through moves and stops [17]. These structures are then used to add information about the trajectory, like the transportation mode during a trip, or the nature of an activity. Such information can, for example, be related to the nature of the moving object, the nature of the mode of transportation, or the visited places. These authors propose especially solutions to enrich raw GPS data indicating the path of a moving object, which result in trajectories encompassing useful information in their description (see Figure 2).
statutes. The conceptual model is based on a relational approach. Later, the model has been modified in order to determine the likelihood of an event occurring under certain conditions in a life trajectory [19]. The conditions of the residential choices are also addressed by [22] with a strong emphasis on the role of job changes occurring in the professional life of individuals. The models proposed in these works are centered on the temporal aspect of residential choices: the focus is more on why people have moved at some place (depending on their life circumstances) than on why they have chosen the place where they have moved to. The aforementioned approaches have clearly laid the conceptual foundations on which we rely for the representation of life trajectories. Our work, however, aim to enrich this representation by a model of semantic trajectories that handles multiple semantic viewpoints and integrates explanatory factors about trajectories, whether these trajectories belong to a single project of life in the minds of individuals, or are drawn from proved observations.
4. A SEMANTIC TRAJECTORY MODEL 4.1 Modeling semantic trajectories
Figure 2 : From raw data to semantic trajectory (From Yan and Spaccapietra [21]) Depending on the scale level of the trajectory, raw data are not necessarily a series of GPS points. Indeed, in the case of life trajectories and when focusing on the residential aspect (or residential trajectory), the essential spatial component is the location of residence, which can be considered at different scales (address, district or, as in our example, the city). This, therefore, implies to consider a sequence of points or areas, indicating the successive residences of a person. When considering the structuration proposed by Spaccapietra [17], the positions of such places of residence, which persist over time, can be considered as "stops" in the life trajectory of an individual.
In order to address the issue of the different viewpoints from which a trajectory is observed, we rely on the concept of point of view1 (see [20]). A viewpoint is a perception among many that an observer has of a studied entity. The point of view consists of a subset of the attributes describing the entity that includes the key attribute – or the set of attributes - which form (s) its identifier as well as a subset of some other attributes. The identifier of the entity is always present in a point of view (or seen from this point of view). Any other attribute can be part of several points of view. According to the most common design of point of view (corresponding to the notion of views in DataBase Management Systems), we consider that the entity, whose trajectory is studied, is unique. The union of the points of view forms a complete picture (all attributes are represented) and coherent (all the attributes have the same value in every point of view) of the entity.
In Figure 2, the moving object is a person (individual, citizen) whose travels and moves are examined at the scale of the day. We understand thanks to some semantics (annotations) added to the trajectory that this person went to work on a bike, and after being back home she drove to the supermarket, information that was not contained in the raw trajectory. More broadly, any information which would make sense or help to describe or explain a trajectory more deeply than raw date (usually a series of triplets (latitude, longitude, time)), should be included in a semantic trajectory model [4].
3.2 Modeling life trajectories The work conducted by Marius Thériault is a reference in the field of the life trajectory modeling. The space-time model for the analysis of life trajectories [18] is based on three different trajectories - residential, family and professional. Each of these trajectories is conceptually modeled by episodes - stable statutes during a time interval - and events altering one or more of these
Figure 3 : Illustration of the notion of point of view Figure 3 shows an example of four viewpoints from which the trajectory of a person can be observed. These viewpoints are adaptable to each person and/or over time, either by adding a new 1
Viewpoint and point of view are considered as synonyms in this work.
attribute in a viewpoint, or by adding a new viewpoint. In our case study, the three viewpoints "residential, "family" and "professional" are essential because they are aspects of life that must absolutely be taken into account to find the explanatory factors behind residential trajectory. Other viewpoints may appear, however depending on the interests of the studied individual, such as leisure activities, which may indeed also influence residential choices. In the following sections, we define the key concepts of our proposal.
4.2 The life trajectory structuration In this work, we combine the approaches of Leibniz and Newton. While the former considers time as a succession of events, the second considers it as a flow measurable in absolute terms [9]. Indeed, we consider the evolution of time as a succession of events, but without renouncing to measure it, using available units (year, day, hour ...). Our approach incorporates considerations carried out by Claramunt and Theriault [9] [8] in their work on time management in geographic information systems. The first problem is to define the temporal granularity adapted to the modeling of life trajectories. According to [12], a chronon is a minimum amount of time, the smallest time stamp unit for an event. In the case of a life trajectory, a relevant chronon may be the day since this unit allows to locate important events in time with a precision that makes sense for an individual (such as the date of her wedding, the birth of a child, etc.). However, at the scale of life it is useful to use a chronon of other granularity, such as month or year, to locate an event for which accuracy in terms of day is impossible (for instance, when the precise date has been forgotten) or not required (for instance, when day accuracy is useless). The model must therefore supports varying degrees of temporal accuracy. Our life trajectory model is composed of episodes which describe the stable state of a person when observed or considered from a given viewpoint. The state of a point of view is considered stable when the values of all its attributes remain unchanged. This modelling choice requires to choose carefully the information to represent (i.e. the relevant attributes), but also to determine the relevant degrees of precision. For instance, regarding the rent of some accommodation, the precise amount might be less relevant than a range of values. This allows to categorize and discretize information (low rent, high rent, etc.) and it has also the advantage of ensuring a certain stability for the value of the involved attributes. It might be indeed pointless for the overall objective (better understand the residential trajectories) to record a new episode simply because the rent has slightly increased. The model is also composed of events that correspond to things that happen and that can be observed according to [1]. The event usually marks the transition from one stable state to another in one point of view (start and end of episodes). Definition of an episode. An episode is the stable state of a person seen from a certain point of view during a time interval. The episode duration corresponds to a time interval between two instants ts and te (as ts < te). Each of these two instants may correspond to an event in the trajectory. Let Epi be an episode of the life trajectory of a person P observed from a point of view i. It is composed of the kP identifier of P and of attributes aj, …, ak, and is valid over a time interval [ts, te]. We note: P.i.Epi(kP, aj, …, ak) [ts, te]
In the example of Figure 1 showing the life trajectory of a person called Leo, an episode valid from 05/20/2000 to 06/30/2004 is noted as follows: Leo.residential.Epi (kP = 123, City="Seattle", Housing_type="Apartment", Rent=[400; 600]) [05/20/2000, 06/30/2004] A non-strict order relation can be established between episodes using algebra of Allen (Allen, 1983). Definition of an event. An event is related to a point of view i for an individual P. It has an identifier kEvt, a name evt_name, and is associated with a chronon t relevant to the observed phenomenon (allowing to locate the event temporally). We note: P. i.Evt (kEvt, evt_name) t For example an event move occurred on 06/30/2004 in the trajectory of Leo and is noted in the residential point of view: Leo.residential.Evt (KEvt = 1, evt_name = "move") 06/30/2004 The evt_name attribute is selected from a controlled vocabulary and describes the nature of the event. An ontology structuring all of the vocabulary specific to each point of view is being defined. Definition of a trajectory. The life trajectory T of a person P consists of the union of the points of view i, each being defined by a set of events and episodes. We note: P.T = {U P.i} for i ∈ {Residential, Professional, Family, ...} A simplified trajectory observed from a single point of view i composed of two events and two episodes with the one still ongoing can thus be noted as follows: P.T.i = {P.i.Evt(kEvt,, evt_name)t1 ; P.i.Epi (kP, aj, …, ak) [t1, t2] ; P.i.Evt(kEvt,, evt_name)t2 ; P.i.Epi (kP, aj, …, ak) [t2, today]} With t1 < t2, t2